Instructions to use Banaxi-Tech/custom-optimizer-model-test with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Banaxi-Tech/custom-optimizer-model-test with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Banaxi-Tech/custom-optimizer-model-test", trust_remote_code=True)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Banaxi-Tech/custom-optimizer-model-test", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Banaxi-Tech/custom-optimizer-model-test with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Banaxi-Tech/custom-optimizer-model-test"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Banaxi-Tech/custom-optimizer-model-test",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Banaxi-Tech/custom-optimizer-model-test

SGLang

How to use Banaxi-Tech/custom-optimizer-model-test with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Banaxi-Tech/custom-optimizer-model-test" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Banaxi-Tech/custom-optimizer-model-test",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Banaxi-Tech/custom-optimizer-model-test" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Banaxi-Tech/custom-optimizer-model-test",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Banaxi-Tech/custom-optimizer-model-test with Docker Model Runner:
```
docker model run hf.co/Banaxi-Tech/custom-optimizer-model-test
```

custom-optimizer-model-test

File size: 1,424 Bytes

51115cb

{
  "parameters": 9968128,
  "architecture": {
    "vocab_size": 8192,
    "hidden_size": 256,
    "num_hidden_layers": 10,
    "num_attention_heads": 4,
    "num_key_value_heads": 2,
    "head_dim": 64,
    "intermediate_size": 768,
    "max_position_embeddings": 4096,
    "rope_theta": 100000.0,
    "rms_norm_eps": 1e-06,
    "tie_word_embeddings": true
  },
  "dataset_id": "HuggingFaceFW/fineweb-edu",
  "dataset_config": "sample-100BT",
  "dataset_revision": "87f09149ef4734204d70ed1d046ddc9ca3f2b8f9",
  "tokenizer_repo": "BananaMind/BananaMind-2-Nano",
  "tokenizer_revision": "c8564d1bd3f6177221ed7e4f63ae5f281a677a1c",
  "optimizer": "Aspect-Cautious Muon",
  "optimizer_description": "Stock Muon plus an aspect-scaled cautious Adam residual on hidden matrices; AdamW on tied embedding and norms",
  "muon_peak_lr": 0.05,
  "residual_peak_lr": 0.0003,
  "residual_aspect_scale_cap": 2.0,
  "residual_cautious_normalization_cap": 2.0,
  "muon_momentum": 0.95,
  "muon_ns_steps": 5,
  "muon_adjust_lr_fn": "original",
  "adamw_peak_lr": 0.003,
  "global_batch": 132,
  "tokens_per_step": 540672,
  "steps": 46238,
  "tokens_seen": 24999591936,
  "target_tokens": 25000000000,
  "world_size": 8,
  "gpu_name": "NVIDIA RTX PRO 6000 Blackwell Server Edition",
  "elapsed_seconds": 5734.559692144394,
  "average_tokens_per_second": 4359461.454424515,
  "estimated_training_flops": 4640871447860871168,
  "seed": 1337
}